E4000 series FSB "wall"

[SoddemFX]

....

[celemine1Gig]

This sounds very interesting.

I know you didn't want to talk about it, but did you measure the resistance between VCCPLL and VSSA on a E4x00 in comparison to the higher E6xx0 Core2 Duo Models? Perhaps they didn't even change filters, but added resistance from VCCPLL to VSS in order lower the voltage, i.e. crippling the max possible FSB that way. Just a thought. I'm sorry if you already did that measurement.

[celemine1Gig]

Hi,

I'm measuring approx 1.6K between VCCPLL and VSS or VCCPLL and VCC, they appear to be isolated on the CPU as suggested in the datasheet but it's hard to say as the chip isn't running. I don't have an E6000 series CPU for comparison.

Was that just an idea or is it more founded?

I was speaking to an engineer today who has a lot of expertise in RF analogue/digital, he said my idea was possible but unlikely.

If this is true then the difference probably lies in the packaging of the CPU and not the CPU die itself. If the CPU die had a weakeness then the VCO would be the first to fall - as the effect is multiplier independant the VCO output is probably good...? Would you agree?

The effect is almost certainly affecting the rest of the PLL components minus the VCO, the worst thing i can think of would be the analogue grounds or VCCPLL and possibly VCCA having a substantial component at the input frequency of the PLL (bus clock). There are a few capacitors present on the E6000 CPU which aren't on the E4000 series.

If we could find the voltage rail being used to supply the analogue voltages and filter at the bus clock frequency it should show wether this is true. A typical 100pF 0603/0805 ceramic hits resonance at approx 400-500MHz which should be good enough.

Any idea where these voltages are supplied from on a P5K? I think they are re-using some other voltage and not dedicated supplies...

Tom

It just occured to me right after having read your posting. BTW, my XEON 3050 ES is measuring 1.134K Ohm from VCCPLL to VSSA.


And yes, I fully agree, or let's better say, I also assume that the modification must be on the package and not the die itself. IMHO, wouldn't make any sense considering the production process.

Unfortunately I cannot help you with your question about the P5K. I never owned one, thus I could only guess.

[celemine1Gig]

I'm not sure if this can be measured in this way tho, and even if it could what would decreasing the resistance of VCCPLL to VSS achieve? Especially as VSS isn't the respective ground to VCCPLL, i think VSSA is.

Tom

I did measure between VCCPLL and VSSA. Just looked it up. Had thought the processor land would be VSS, but it was VSSA indeed. Although it made no difference at all. The same 1.134k on VSS, too.

I'm not sure either if it can really be measured that way.
As we have no exact information we will most likely be guessing to a certain degree anyway.

Perhaps some more poeple could do some measurements. Then we would see quite fast if there is a trend or if it's just a random difference.

[celemine1Gig]

VSSA and VSS are probably connected on the motherboard, god knows how they're related on the CPU

I think it could be noise on one of the analogue voltages / grounds. I will try a few things, if anything works i will let you know.

Yes, i anyone wants to measure whatever across a range of CPU's it wouldn't hurt at all

Tom

Yes, be sure to keep up the good work. I'd really like to be more helpful, but unfortunately I doubt that I can be.

Just one more rrandom thought: If they intended to increase the noise on the analogue voltages to hamper the FSB clocking, then wouldn't it be possible that they didn't really isolate VSSA from VSS? Still just guessing.

[celemine1Gig]

I don't think it was ever "sabotage"

Thinking how i would do it - i'd not deliberatly compromise the design i'd just reduce the non essential components, if it only needs to run at 200MHz why equip it with analogue systems capable of 333MHz...?

If a motherboard maker produces a budget board, it's designed to run the best it can, if it's sold to the low end they just don't fit the capacitors or inductors needed to handle 150A if it only needs 80A. That's what i think anyway

Tom

//F*ck me - there's the pads under the packaging substrate to take more capacitors! 4x which looks 0402 size, 2x which look 0603 - they're buried under the solder resist, you have to tilt the CPU to see them - the pads are there!

Now that sounds good. And concerning the "sabotage" part: I think we both mean to say the same thing. It's just a bit more difficult for me to express as english is not my mother-tongue. IMHO, it is sabotage in a certain way. I mean they really wanted it not to be capable of high FSB clocks, but it's not done the dirty way that sabotage usually looks like, but the sophisticated, professional, engineering way.

Ok - What will take off the resist other than a scalpel?

How about an angle-grinder? Or do you intend to still use the CPU after the modification?

[Warship]

subscribed

[sang]

VSSA and VSS are probably connected on the motherboard, god knows how they're related on the CPU

I think it could be noise on one of the analogue voltages / grounds. I will try a few things, if anything works i will let you know.

Yes, if anyone wants to measure whatever across a range of CPU's it wouldn't hurt at all

Tom

VccPLL to Vssa/Vss - 60kohms

[img=http://img231.imageshack.us/img231/5237/img1910asj0.jpg]
[img=http://img235.imageshack.us/img235/7445/img1912aqx3.jpg]

And i dont think its the noise to be blame for.
Ive isolated/separated the VccPLL from the Southbridge voltage on my DFI 965P-S and after feeding the VccPLL with a personal excellent noise filtered "maximizer" (built on a lm317a ),all i have got were just 15Mhz on BUS over my fsb wall. But just was the couse of extra voltage(like 1.95v vs. 1.8v max default)

[celemine1Gig]

VccPLL to Vssa/Vss - 60kohms

[img=http://img231.imageshack.us/img231/5237/img1910asj0.jpg]
[img=http://img235.imageshack.us/img235/7445/img1912aqx3.jpg]

And i dont think its the noise to be blame for.
Ive isolated/separated the VccPLL from the Southbridge voltage on my DFI 965P-S and after feeding the VccPLL with a personal excellent noise filtered "maximizer" (built on a lm317a ),all i have got were just 15Mhz on BUS over my fsb wall. But just was the couse of extra voltage(like 1.95v vs. 1.8v max default)

Please fix the links to your pictures.
And some more detailed description of your mods would be really nice.

[sang]

Hi Sang,

I agree with you.

Are you saying that on your DFI965-P VCC_PLL is supplied from the southbridge voltage?.

like most of the 965 mobos,Yes.

When you say you isolated VCC_PLL from the southbridge and used your own linear supply, do you mean that you physically disconnected the two and fed just VCC_PLL from your own linear supply?

Exactly what i have done.

Now the PLL mod is off since im playing with another board,but ill put it back monday evening and get back with some pictures.
Honestly,my hopes were greater than 15mhz up over normal fsb wall. Even with cpu on -135C i couldnt pass 590 bus on modded DFI P965 ,though my P5K Vanilla,identical test/hardware condition goes for 620.

Ill try to make a compare very soon,checking for signal clarity of VccPLL with my mod on/off over the DFI P965-S. Got an osciloscope from a friend and maybe this will make things more visible

[sang]

Please fix the links to your pictures.
And some more detailed description of your mods would be really nice.

oops, sorry for that.
Ill fix them monday,got the exif on my hdd at work.
The mod is pretty simple.
D23 - follow the trace
Ill get back with some pictures.

[celemine1Gig]

...
The mod is pretty simple.
D23 - follow the trace
...

I've already done that on my AW9D. But I only changed the original resistance to increase VCCPLL.
Still some more pics of your mod would be nice anyway. Always nice to see what ideas other people come up with. Modding is a field where creativity counts.

[sparkY1000]

WOW! I cant wait to see you results Tom!

Incidentally what multi are you running with at 360MHz FSB?

The SMD caps I was looking at for you were the X7R ones, but most of the companies I tried wanted a trade customer buying 1000's not me buying 50 or so lol!

[Dualist]

Just had search at RS components and they do SMT X7R's if they are the right ones and they sell them in batches of 10
Hope this is of some use.

[webstien]

Subscribed, interesting thread

[sang]

DFI 965-s PLL Mod

Somehow kinda useless as you need an external linear supply (1.5v ->3v).
Anyway,here is it:




my E6700
before: fsb wall (@1.80v-SB1.5v) at 545,aircooling
after the mod: fsb wall (@2.15v PLL) at 561, aircooling

osciloscope tests for signal quality soon
also,if anyone interested for Vcore on DFI P965-s/non S, just say ..

[MaSell]

any news?

[STEvil]

Looking forward to how this progresses

Just out of curiosity i've noticed some "empty" pads on the bottom of many cpu's which appear to be debugging pads and etc, could be possible that putting power or grounding these might open or close off features maybe, no?

[celemine1Gig]

Nice! Best of luck from me! Hope the CPU still works.

[sparkY1000]

I would try it in the Assrock first just to make sure it works before risking the P5K dude! :'(

[STEvil]

got an e4300 and asrock board here but the board is limited to 300fsb absolute maximum..

[dinos22]

Hi MaSell,

I have the capacitors soldered to the CPU but i'm too affraid to put it in the board at the moment.

It was soldered under a microscope, i was originally going to try to do it but in the end i got help from someone more experienced, it took us about 3 hours all together. The gaps between the pads and the gound plane on the packaging substrate is tiny, it has to be in the region of 0.1mm. Soldering the 0402's was a nightmare it took me and another guy about 3 hours, we fitted 4 x 220pF 0402's and 2 x 10uF 1206's both X7R. There are no visual shorts under the microscope but you can't measure any shorts anyway - it's try and see in the board.

I'll be putting it in the P5K this weekend - fingers crossed it still works at all. Even if it goes pop i've got some nice-ish pictures of the CPU on my camera back home

I did what Sang did as well (i was bored), but it didn't gain me anything realy. Off a 5v rail it lost me about 5MHz, on the battery it gained me about 2MHz - woooooo. VCC_PLL on these E4000's must be a b*tch for noise, draws next to no power tho, ran for a good few mins on the PP3 dropping from 9v to ~2v over a linear reg



Anyway - that was something else i wanted to play with but with the CPU mods if it goes boom or works i'll let you know anyway

Tom

[Antinomy]

Just out of curiosity i've noticed some "empty" pads on the bottom of many cpu's which appear to be debugging pads and etc, could be possible that putting power or grounding these might open or close off features maybe, no?

Better not doing that these are not (as I understood you) debugging pads, but the voltage pads. Some are Vcc, some Vss or whatever. If you put power on a power rail it can make some smoke
The debugging is in JTAG interface, and the pads are in the LGA775. Some of the empty pads can be reserved for caps, but not used (that what this thread os about, as I read).
Can you make a foto of them?

[STEvil]

http://www.xtremesystems.org/forums/...7&postcount=15

The smaller dots (and one square) around the inside of the square left in the center of the LGA pads in that pic

[Antinomy]

Better check the resistance to Vss and Vcc on these. Maybe it's just and outpit of them. Because they do not look like pads.
We were thinking of the FSB wall problem on ROM.by. But the only two ideas are modification of voltages (Vcca and et cetera) and the filter on the bottom. I think, a comparision of the Celeron 4xx, E21xx, E4000, E6000 and Quad should be made to see differences.